The effects of skewed theta distributions on indices of differential item functioning (DIF) were studied, comparing Mantel Haenszel (N. Mantel and W. Haenszel, 1959) and DFIT (N. S. Raju, W. J. van der Linden, and P. F. Fleer) (noncompensatory DIF). The significance of the study is that in educational and psychological data, the distributions one deals with are sometimes not normal, and that could translate into a skewed distribution. Monte Carlo procedures were used to generate the data under normal and skewed conditions for 1,000 and 500 examinees each in focal and reference groups and a test length of 40 items. The thetas for the examinees were drawn from normal, positively skewed, and negatively skewed distributions. Item and ability parameters were estimated though the BILOG computer programs. The effectiveness of each DIF index was determined by examining the true positive ("hits") and the false positive rates. The Mantel Haenszel, Lord's chi square, and noncompensatory DIf detected uniform DIF items very well. Precision of detection increased as sample size increased with all methods. All three methods tended to have difficulty detecting nonuniform DIF items. Precision declined for all methods at the extreme skew, but moderate skews do not affect detection much at all. It is recommended that researchers use a combination of DIF detection methods as seems appropriate to the conditions. (Contains 11 tables and 23 references.) (SLD)